The local dashboard
Open lean-ctx dashboard to inspect the runtime’s available local activity and metrics. Its visibility follows the operations your integration records.
Explore this capabilityEvidence & Outcomes
The record first.Then the result.
LeanCTX is the Context Gateway for AI systems. See what LeanCTX delivered to the model, which rule applied and what was withheld. Usage and estimated cost come from the same record.
The approach
One record. Delivery, rules and usage.
LeanCTX records what the supported integration delivered, the configured checks that affected delivery and the usage behind it. Use that record alongside tests or review to judge whether a workflow works better.
Inspect
Inspect the delivered input.
Use recorded source references, plans and token measurements to inspect supported operations. Available fields depend on the operation and runtime.
- Source
- Identify the file, reference or plan behind the operation.
- Delivery
- Inspect recorded output and available token evidence.
- Scope
- Receipt fields depend on the operation and runtime surface.
Trace & replay
Follow the evidence back.
Trace a receipt to its source or plan and revisit the state still available. Reproducing a model answer also depends on the host, model and request settings.
- Linkage
- Connect an operation with its recorded source and plan references.
- Availability
- Missing source or expired state limits what can be revisited.
- Reproduction
- A context record alone does not guarantee the same model answer.
Usage & economics
Usage is part of the record. Not the whole bill.
Start with measured usage. Keep estimated model cost separate from observed charges and compare equivalent tasks. The captured example shows input reduction for one read, not a promise about total spending.
- Input
- Tokens before and after the selected preparation.
- Cost basis
- Identify estimates, observed charges and unavailable values explicitly.
- Quality
- A separate result established by task checks.
src/lib/roi.tssignatures · o200k_base
82.9%
less input for this read
2,127 tokens removed
A signature view omits implementation details. This is an input measurement, not a cost or quality result.
Outcome & feedback
A completed call is not a completed task.
Evaluate against a documented baseline and explicit acceptance criteria. A provider response, a caller-reported success and a verified outcome are different evidence. Use accepted results to inform the next context and routing decision.
- Baseline
- The same task and a documented input configuration.
- Outcome
- Tests or review criteria that matter for the actual task.
- Comparison
- Include failed attempts and retries when comparing cost per accepted task.
In your workflow
From one operation to your workflow.
Use the right inspection surface for each question. A dashboard, a source receipt and an evaluation report reveal different parts of the work.
Cost attribution
Distinguish model usage, cached input, output and repeated attempts. Preserve the basis of estimated or observed cost; an unknown charge is not zero cost.
Explore this capabilityAn evidence-based feedback loop
Compare equivalent tasks against a documented baseline, including failures and retries. Feed verified outcomes back into later decisions without turning a caller’s success label into proof.
Explore this capabilityQuestions & answers
Measurement questions.
Do fewer input tokens mean the same percentage off my bill?
No. Model prices, cached-input rates, output tokens and the number of calls also affect cost. The recorded demo measures the input reduction for one file read.
Does a receipt prove answer quality?
A receipt records an operation. Quality needs its own evidence: tests, source checks or a defined human review. Use both when deciding whether a workflow is effective.
Start with LeanCTX
Measure your own workflow.
Choose a representative task and a baseline. Compare the input, then check the outcome against your own criteria.
Local-first. Model-agnostic. Your context.